Revealing cosmological fluctuations in 21cm intensity maps with MeerKLASS: from maps to power spectra

TL;DR

MeerKLASS demonstrates the viability of single-dish 21cm intensity mapping with a multi-dish array by developing a robust end-to-end pipeline from calibrated maps to power spectra. Through PCA and multiscale foreground cleaning, signal-loss correction via transfer functions, Cartesian regridding, and careful covariance treatment, the work achieves robust cross-correlation detections with optical surveys and progresses toward auto-correlation measurements. Forecasts show that a wide-area UHF-band survey will deliver competitive constraints on the growth rate, BAO, and primordial non-Gaussianity, especially when combined with multi-tracer data from Rubin LSST and future SKAO surveys. The study highlights the critical role of forward modeling, systematics control, and innovative analysis techniques in unlocking the cosmological potential of 21cm intensity mapping at low to intermediate redshifts, with clear pathways toward the SKAO era.

Abstract

Mapping the integrated 21cm emission line from dark matter-tracing neutral hydrogen gas is the primary science goal for MeerKLASS (MeerKAT's Large Area Synoptic Survey). Prior to the arrival of MeerKAT, this intensity mapping technique had only been tested on a couple of pre-existing single-dish radio telescopes with a handful of observational hours with which to make early pioneering detections. The 64-dish MeerKAT array, precursor to the Square Kilometre Array Observatory (SKAO), can scan the sky in auto-correlation mode and perform intensity mapping across large sky areas, presenting the exciting potential for a wide-sky (${\gtrsim}\,10{,}000\,{\rm deg}^2$) spectroscopic survey across redshift $0.4\,{<}\,z\,{<}\,1.45$. Validating the auto-correlation (or single-dish) mode of observation for a multi-dish array and developing the analysis pipeline with which to make unbiased measurements has presented major challenges to this endeavour. In this work, we overview the advances in the field that have facilitated a robust analysis framework for single-dish intensity mapping, and review some results that showcase its success using early MeerKLASS surveys. We demonstrate our control of foreground cleaning, signal loss and map regridding to deliver detections of cosmological clustering within the intensity maps through cross-correlation power spectrum measurements with overlapping galaxy surveys. Finally, we discuss the prospects for future MeerKLASS observations and forecast its potential, making our code publicly available: https://github.com/meerklass/MeerFish.

Figures (10)

• Figure 1: (left): Accuracy of reconstructed foreground cleaned power spectra relative to the foreground-free Hi-only data ($P_\textrm{Hi}$) from simulated intensity maps. The transfer function, $\mathcal{T}(k)$, is constructed using \ref{['eq:TF']} and 100 lognormal mocks, independent of the main $N$-body simulation assumed as the observational data. The shaded bands show the rms over these 100 mocks. A mild ($N_\text{fg}\,{=}\,8$, blue line) and more aggressive ($N_\text{fg}\,{=}\,12$, red line) foreground clean are shown. Dark-thick (light-thin) green horizontal lines indicate sub 1% (5%) accuracy regions of the reconstructed power spectrum. (right): The shape of the transfer. Solid lines indicate noise-free simulations, thin dashed lines are where dominant white noise with rms $\sigma_\text{n}\,{=}\,1\,\text{mK}$ is added to the simulated observations.
• Figure 2: (left): Pixelisation and Cartesian gridding processes necessary for intensity mapping analysis in Fourier-space. The background image represents a continuous fluctuation field. Intensity mapping surveys will discretise calibrated observations of the continuous field into broad sky pixels (blue dashed boundaries in insert). This provides the 3D voxel array in sky coordinates (R.A., Dec, $\nu$). Using random sampling particles, the sky voxel intensities can be regridded onto a 3D Cartesian grid (red boundaries) in comoving space ($l_\text{x}$, $l_\text{y}$, $l_\text{z}$ in $h^{-1}\text{Mpc}$). (right): Accuracy of measured power spectrum from regridded fields relative to input model (black lines) for the two simulation versions (Cubic voxels and Fine channels). We use NGP interpolation for with no compensation (blue lines) and with compensation using \ref{['eq:Compensated']} and \ref{['eq:W_ngp']} (orange lines). The $1\sigma$ scatter from the 100 simulation realisations is shown by the shaded regions.
• Figure 3: Cross-correlation power spectra between MeerKAT L-band intensity maps and optical galaxy surveys at $z\,{\sim}\,0.43$. Blue-circles show the initial pilot survey detection with WiggleZ galaxies Cunnington:2022uzo (hollow markers indicate negative power), orange-diamonds the re-analysis of this data with improved foreground cleaning Carucci:2024qpm, and the green squares is the deep-field cross-correlation with GAMA MeerKLASS:2024ypg. Error bars show $1\sigma$ uncertainties estimated using different methods in each detection. The dashed lines are the corresponding model matched by colour. We refer the interested reader to the dedicated papers for details on these subtle distinctions.
• Figure 4: Two-dimensional cross-power analysis between the MeerKLASS L-band deep-field intensity maps and GAMA galaxies. The panels show (from left to right) the measured cross-power (with signal loss correction), the computed foreground transfer function, the estimated signal-to-noise, and the number of modes contributing to each $(k_\perp, k_\parallel)$ bin. The orange lines indicate the chosen $k$-cuts used to exclude poorly constrained regions of Fourier space from the spherically averaged $P(k)$ measurement (green results in \ref{['fig:crossPks']}), and the grey contours show the $k$-bin boundaries chosen for this $P(k)$.
• Figure 5: The Hi auto-power spectrum for the MeerKLASS L-band deep-field from MeerKLASS:2017vgf. The red solid line is the noise estimate from measuring the average power of 10 Gaussian random fields with an RMS expected to be consistent with the thermal noie level. The grey dotted line is the predicted Hi power spectrum model and the black dashed line is the combination of this Hi model plus the thermal noise estimate.